modelling optimal allocation of resources in the context of an...
TRANSCRIPT
RESEARCH ARTICLE
Modelling optimal allocation of resources in
the context of an incurable disease
Betty Nannyonga1☯*, David J. T. Sumpter2☯
1 Department of Mathematics, School of Physical Sciences College of Natural Sciences, Makerere
University, Kampala, Uganda, 2 Department of Mathematics, Center for Interdisciplinary Mathematics
Mathematics Institution, Uppsala University, Uppsala, Sweden
☯ These authors contributed equally to this work.
Abstract
Nodding syndrome has affected and led to the deaths of children between the ages of 5 and
15 in Northern Uganda since 2009. There is no reliable explanation of the disease, and cur-
rently the only treatment is through a nutritional programme of vitamins, combined with med-
ication to prevent symptoms. In the absence of a proper medical treatment, we develop a
dynamic compartmental model to plan the management of the syndrome and to curb its
effects. We use incidence data from 2012 and 2013 from Pader, Lamwo and Kitgum regions
of Uganda to parameterize the model. The model is then used to look at how to best plan the
nutritional programme in terms of first getting children on to the programme through out-
reach, and then making sure they remain on the programme, through follow-up. For the cur-
rent outbreak of nodding disease, we estimate that about half of available resources should
be put into outreach. We show how to optimize the balance between outreach and follow-up
in this particular example, and provide a general methodology for allocating resources in
similar situations. Given the uncertainty of parameter estimates in such situations, we per-
form a robustness analysis to identify the best investment strategy. Our analysis offers a
way of using available data to determine the best investment strategy of controlling nodding
syndrome.
Introduction
In the northern part of Uganda, thousands of children have fallen ill with a little known, incur-
able nodding disease syndrome, NDS. Communities are panicking and losing hope as neither
the cause nor a cure are known [1]. The affected Ugandan districts include Kitgum, Lamwo,
Pader and Gulu. The nodding syndrome emerged in Sudan in the 1960s [2], although it was
first reported in 1962 in secluded mountainous regions of Tanzania [3]. Since then, it has been
reported in small regions in South Sudan, Tanzania and Uganda [4, 5]. However, the connec-
tion between that disease and the nodding syndrome was just made recently [5].
Nodding disease affects young children between the ages of five to fifteen. It is mentally and
physically disabling, eventually fatal. As of 2013, there were more than 3,094 cases and 170
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 1 / 14
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
OPENACCESS
Citation: Nannyonga B, Sumpter DJT (2017)
Modelling optimal allocation of resources in the
context of an incurable disease. PLoS ONE 12(3):
e0172401. https://doi.org/10.1371/journal.
pone.0172401
Editor: Delmiro Fernandez-Reyes, University
College London, UNITED KINGDOM
Received: July 1, 2016
Accepted: February 3, 2017
Published: March 13, 2017
Copyright: © 2017 Nannyonga, Sumpter. This is an
open access article distributed under the terms of
the Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: All relevant data are
within the paper and its Supporting Information
files.
Funding: BN acknowledges funding from the
Faculty for the Future and the Center for
Interdisciplinary Mathematics, Mathematics
Institution, Uppsala University, Sweden. The
funders had no role in study design, data collection
and analysis, decision to publish, or preparation of
the manuscript.
Competing interests: The authors have declared
that no competing interests exist.
deaths in northern Uganda [6]. Between 2012 -2013, the government of Uganda set up out-
reach and screening centers at health facilities but due to deep poverty, many families could
not afford to pay for transport to these facilities. Children affected by NDS are characterized
by complete and permanent stunted physical growth [7]. The brain is also stunted, leading to
mental retardation of the symptomatic individual. Due to pathological bobbing of the head,
the disease is named nodding syndrome. The nodding was discovered to be an atonic seizure
[8], which often begins when children eat or feel cold. The seizures are brief and stop when the
child stops eating or feels warm again and they can manifest with a wide degree of severity.
When severe, the affected child can collapse, which may result in further injury [9], such as
falling into the open fire. Although inconclusive, there is documented association of the dis-
ease with malnutrition and onchocerciasis. NDS is a progressive fatal disorder with a duration
of about three years or more [10]. A small fraction of the symptomatic survive, and the major-
ity dies [7].
Nodding disease started spreading across the Sudan-Uganda border in 2009 where more
than 2000 cases [3] were reported in Uganda’s Kitgum district [4]. By the end of 2011, NDS
outbreaks were concentrated in Kitgum, Pader and Gulu [11]. The spread and manifestation
of outbreaks may be made worse by the poor health care of the region [10]. Children suffering
from nodding disease syndrome are exposed to poverty and malnutrition, creating a fresh dis-
tress to the already suffering region [12], and exposing children to otherwise preventable
deaths. The recent results from CDC [13] report crystals found in victims’ brains. Still, there
are no answers to why the illness attacks only children, and in most cases, some and not all in a
family. Further, there is no answer of whether there is a genetic predisposition. The impact of
NDS has been highly destructive and health officials are left mystified as to cause and cure.
Children living under the poorest conditions with insufficient food, no safe drinking water, or
indecent housing are most susceptible to this condition. In addition, the affected children are
often neglected by their parents [14] and barred from schools [15].
The outreach and screening centers set up by the Uganda government provide nourishing
foods and medication for symptoms to the patients. Each month, patients are examined to
determine any improvement in the level of malnutrition, frequency and severity of seizures.
The nutritional programme is aimed to minimize the symptoms such as occurrences of sei-
zures. However, resources are extremely limited. The research question we address is how to
balance the use of resources in getting children on to a nutritional programme and symptom
preventing medicines in the first instance through outreach, against keeping them on this
treatment programme once they have started, through follow-up. In many Uganda health con-
texts, outreach means that services that are not easily available are taken nearer to the popu-
lace. Once these services are set up, getting the children to go to the centers involves
community effort through local councils visiting their respective villages and urge the next of
kin to take them there. Follow-up is when children return the following month for check up
and additional food supplements where required. In the absence of any cure in the foreseeable
future, we look at how to best utilize the limited available resources to minimize the impact of
the disease. We address this problem using a mathematical model.
Mathematical models have been used for optimization of disease prevention [16–19] and
epidemic control [20–26]. The models enable more efficient searches for desirable control
strategies by considering all strategies simultaneously. The results direct the models in generat-
ing detailed predictions of potential future outbreaks. An optimization model on foot-and-
mouth disease [27], was used to evaluate alternative control strategies that minimized the total
regional cost for the entire epidemic duration, given disease dynamics and resource con-
straints. A similar model using optimal control theory was developed for dengue fever [28] to
analyze the optimal strategies for using these controls, and respective impact on the reduction/
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 2 / 14
eradication of the disease during an outbreak in the population. A model was designed to
investigate problems of disease prevention and epidemic control, in which two sets of deci-
sions namely vaccinating individuals and closing locations were optimized with the goal of
minimizing the expected number of symptomatic individuals after intervention [29]. Similar
to our objective, the goal of the paper was to formulate mathematical optimization models, to
identify which subset of individuals should be vaccinated and which locations to close in order
to minimize the expected number of symptomatic individuals. Results of such models have
proved beneficial in decision and policy making on management of disease outbreaks and
prevention.
In our model, three states are used: the symptomatic, treated and stunted. The symptomatic
children that have not been exposed to any form of treatment or food supplements. The
treated, are those that have received medication for symptoms of nodding syndrome including
malnutrition, seizures, and poor growth. The stunted are the children whose growth has been
inhibited due to nodding disease. These have either never received any medication or food
supplements, or have dropped out of treatment. Using the model, we determine the optimal
percentage of investment in treatment vs follow-up, with the overall objective of minimizing
stunted growth of the affected children.
Available data
The data used in this study was collected from Pader and Lamwo districts in the northern part
of Uganda between 2012 and 2013 by the Ministry of Health Uganda. Outreach centers were
established in the districts and serviced villages in respective parishes and sub-counties. A
team consisting of twenty (20) personnel from the Ministry of Health visited children once
every month in ten (10) different centers where each health official worked for six (6) hours
from 10:00 am to 4:00 pm, giving a total of 1200 man hours. In order to operationalize this
data to use it in the model, for each the 994 patients, we recorded
1. Age at first visit;
2. Date of initial onset;
3. How many times the patient accessed medical treatment again within six months;
4. The home villages of each of the patients.
Children were brought into the health centers by their parents or next of kin who were
informed of the study and verbal consent obtained to treat, evaluate and follow-up. On their
first evaluation, history and physical assessment of the affected children was done and
recorded. The healthcare officials used the gathered information to determine the severity of
the syndrome and level of malnutrition for each patient. In severe cases of acute malnutrition,
food supplements including ready-to-use- therapeutic foods (RUTF) were given to control
undernutrition and prevent death. Sodium valproate (INN), an anticonvulsant used in the
treatment of epilepsy, was given to control the atonic seizures. On subsequent visits, individu-
als were assessed on whether they were taking the drugs consistently or if they defaulted. The
data used in this study was authorized by the Ministry of Health Uganda who approved this
retrospective study. The records used in the study were anonymized and de-identified prior to
analysis.
Model formulation and analysis
We consider a population of children with and without nodding syndrome. Children with the
syndrome are divided into three sub groups depending on whether they seek medication and
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 3 / 14
follow-up or not. Subgroup I gives the total number of children, who were previously healthy,
but currently with malnutrition, reported head nodding and seizure episodes. Subgroup T are
the children enrolled by the healthcare officials to receive monthly medication for displayed
symptoms and/or food supplements in case of malnutrition. Children in R have stunted
growth, classified based on visible growth impairment that was determined using the current
height as compared to previous determinations and body weight corresponding to current
age. The children are impaired due to absence of or fall-out from treatment and malnutrition.
More details are shown in Fig 1.
Nodding syndrome has been reported to exist in children aged between 5–15 years [2, 30,
31]. From the data, the minimum age at which a child was recorded to have nodding syn-
drome symptoms was 2 and the maximum 27. However, out of the 994 individuals recorded,
947 (95%) are between the ages of 5–16. Therefore, the age group of 5–16 is used to estimate
the rate at which children develop nodding syndrome. A child enters subgroup I at a rate r,
when they start to display nodding syndrome symptoms such as head bobbing, seizures and
malnutrition. Prior studies have given the prevalence of the nodding syndrome in Kitgum dis-
trict to be 12 per 1000 [30, 31]. We shall use the same value for Pader and Lamwo since they
are in the same geographical region sharing borders. In order to transform this estimate into a
rate per month we divide by the age range over which the disease can occur and the months in
a year, i.e. r = 12/(1000 × 11 × 12) = 0.00009. The rate of development of nodding syndrome is
a function of the total number of children NC, in each district. In Kitgum, 45,800 children were
reported in the age range of 5–15. The total population of Kitgum is 247,800, implying that
45,800/247,800� 18.48% is the percentage of susceptibles. We estimate NC for Lamwo and
Pader by taking 18.48% of their total populations. The population of Lamwo is 171,300, and
231,700 for Pader [32], giving the estimated number of children in these two districts as NC =
0.1848 × (171,300 + 231,700)� 74,474. Therefore, the number of incident cases that develop
nodding syndrome per month rNC = 6.7027.
The interest of this study is to minimize the number of children that develop stunted
growth. We can do this by moving children from compartment I to T, and prevent move-
ment to R. Children from subgroup I can move to subgroup T if they seek treatment, or R if
they do not seek medication for the symptoms. Movement from I to R occurs when children
develop stunted growth due to lack of medication for displayed symptoms. We assume that
on average, a child becomes removed due to NDS after six months from onset. Therefore,
the untreated development rate of nodding syndrome γ = 1/6. According to the data, of the
Fig 1. Flow diagram for the dynamics of nodding syndrome.
https://doi.org/10.1371/journal.pone.0172401.g001
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 4 / 14
994 recorded cases, 703 accessed medication and nutritional supplements at least twice in
six months. Therefore, the proportion that did not access medication at least twice in six
months is 291/994 = 0.2928. This gives the rate of fall-out from medication per month as
β = 0.2928/2 = 0.1464.
The movement from I into T depends on whether children seek medication, which in turn
depends on the outreach effort by healthcare personnel. Movement from T to R depends on
the efficiency of follow-up by medical personnel. In the absence of follow-up activities, chil-
dren can start to display nodding symptoms. Resources for outreach and follow-up are limited,
and medication for existing cases is prioritized. Professionals from the Ministry of Health set
up ten (10) health centers in each district, where they visited each center and attended to the
children once every month. The total number of hours they invested per month can be esti-
mated from the data using the number of hours each healthcare official used on each visit at
each center. Since 20 healthcare providers visited the districts for 6 hours a day (from 10:00am
to 4:00pm), 10 times per month, the total available man hours per month is H = 20 × 6 ×10 = 1200.
Of the total available man hours per month H, a proportion is used for treatment of existing
cases. We estimate the time taken to handle one child identified with nodding syndrome and
in treatment to 30 minutes [33, 34]. Thus each treatment takes τ = 30/60 = 0.5 hours. There is
one visit per treated case per month and therefore the total hours per month used for treat-
ment is τT. As a result, the total available hours for outreach or follow-up is H − τT.
Of the total remaining hours after treatment H − τT, a proportion can be invested in out-
reach and some in follow-up. Let pt be the proportion of hours dedicated to outreach activities
for getting children in treatment, and pf for the proportion of hours dedicated to follow-up,
with pt + pf = 1. Then, the total number of hours used for outreach that gets children into treat-
ment is pt(H − τT) per month, with pf(H − τT) hours used for follow-up. We do not set pt at
this point, but rather aim to find a value of pt that gives the best strategy to invest in getting
and keeping children into treatment.
In order to optimize pt and pf, we have to make modelling assumptions about how effort is
translated into success. For the transition rate from I to T the rate can be anything from 0 if no
hours are invested in outreach to 1, for a situation where healthcare is widely available, and
symptomatic children enter treatment within one month of displaying symptoms. We set a
constant α = 1 to give the saturation level of outreach when at its most effective. The transition
function of children from state I to T can then be written as
aptðH � tTÞ
Aþ ptðH � tTÞ
so that it increases with investment in outreach. τ is the time taken to handle one child per
visit, set to 30 minutes [33, 34] and A is the half saturation constant, which is the point where
children from at least 50% of the villages received medication. This value can be estimated by
taking 50% of the total number of villages in the data (211), giving A = 50% × 211 = 105.5.
For the transition rate from T to R, a child would fall-out from treatment and develop
stunted growth at a rate β. The fall-out rate can be close to 1 when more hours are invested in
getting children into treatment and close to 0 if follow-up care of at least twice in six months is
available. The constant β = 0 signifies the success and effectiveness of follow-up care. The tran-
sition function of children from state T to R is given as
bg 1 � expð� tT=ðpf ðH � tTÞÞÞ� �
;
and is large if investment in follow-up is small.
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 5 / 14
With the descriptions of I, T and R, we can assume that δR > δI > δT since children in T are
receiving medication and follow-up and therefore die at a slower rate than I, whereas children
in R are stunted and sometimes brain dead due to the numerous seizures and more often die
faster from other accidents such as falling in open fires during abrupt seizures. With outreach,
medication and follow-up, symptomatic children recover within a period of three years
(d = 3 × 12 = 36 months) or more, or die from the symptoms [10]. With the assumption that
36 months is the time an individual should recover or die from the nodding syndrome, we set
the disease induced death rates for I to δI = 1/36, for T to δT = 0.5 × δI, and R to δR = 1.5 × δI.
These modification parameters are guesses that respectively inhibit or enhance death rates
with or without medication. We summarize all the parameters discussed in Table 1, and give a
set of equations describing the process in Eq (1).
From the above descriptions and assumptions, we can now write the system of equations
describing the system as follows:
dIdt¼ rNC �
aptðH � tTÞAþ ptðH � tTÞ
I � gI � dI I;
dTdt
¼aptðH � tTÞ
Aþ ptðH � tTÞI � bg 1 � e
� tTpf ðH � tTÞ
0
B@
1
CA � dT T;
dRdt¼ gI þ bg 1 � e
� tTpf ðH � tTÞ
0
B@
1
CA � dRR:
ð1Þ
To analyze the model in Eq (1), we start by transforming it using the following approximations:
If the proportion of investment in getting children in treatment is very small, (i.e. small pt(H −τT)), αpt(H − τT)I/(A + pt(H − τT)) can be approximated to αpt(H − τT)I/(A). Small investment
in getting children in treatment implies big investment in follow-up. Therefore, when pt(H −τT) is very small, pf(H − τT) is very big. At this point, βγ(1 − exp(τT/(pf(H − τT)))) tends to
Table 1. Parameter estimates from Uganda data.
Parameter Description Dimensions
r Development rate 0.00009 / month
NC Number of children 74,474
γ Stunted growth development rate 1/6/month
β Medication fall-out rate 0.1464 /month
H Total man hours per month 1200 man hours/month
τ Number of hours per visit per child 0.5 hours/visit/month
pt Proportion in treatment [0,1]
pf Proportion in information [0, 1 − pt]
α Rate at which treatment starts in ideal situation 1/month
A Half saturation constant 105.5
d Duration of nodding syndrome before recovery/death 36 months
δI Disease induced death rate of symptomatic 1/36 /month
δT Disease induced death rate of those on treatment 0.5/36 /month
δR Disease induced death rate of the stunted 1.5/36 /month
Parameter estimates for Lamwo and Pader.
https://doi.org/10.1371/journal.pone.0172401.t001
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 6 / 14
βγτT/(pf(H − τT)). Since pt + pf = 1, we make the substitution pf = 1 − pt and approximate
Model (1) to
dIdt¼ rNC �
aptðH � tTÞA
I � gI � dI I;
dTdt
¼aptðH � tTÞ
AI �
bgt
ð1 � ptÞðH � tTÞT � dTT;
dRdt¼ gI þ
bgt
ð1 � ptÞðH � tTÞT � dRR:
ð2Þ
In terms of the treated T�, the equilibrium points are given by
I� ¼A
aptðH � tT�Þbgt
ð1 � ptÞðH � tT�Þþ dT
� �
T�;
R� ¼1
dR
gAaptðH � tT�Þ
� �bgt
ð1 � ptÞðH � tT�Þþ dT
� �
þbgt
ð1 � ptÞðH � tT�Þ
� �
T�;
0 ¼ rNC �AT �
aptðH � tT�ÞaptðH � tT�Þ
Aþ ðgþ dIÞ
� �bgt
ð1 � ptÞðH � tT�Þþ dT
� �
:
ð3Þ
Due to complexity of the model, analysis can be done numerically to obtain mathematical con-
clusions by simulating the steady states for different values of pt 2 [0, 1]. We assume that once
the children with nodding syndrome are identified, they all opt for treatment within a month.
This assumption helps us to obtain the point for optimal investment in the benefit of treatment
and information. The optimal value of pt corresponds to the best strategy of investment and
gives the minimum number of patients that develop stunted growth or equivalently the maxi-
mum number enrolled for medication.
Fig 2 (blue line) shows that at the beginning of investment in outreach (about 2%), there is
no significant increase in the number of children enrolled in treatment. However, the red line
representing the stunted children shows a slight fall in the number of stunted growth followed
by a sudden jump, and then later to a steady decrease to a minimun value attained when opti-
mal investment is achieved. With further outreach, the blue curve shows the point at which the
number of children on treatment T increase to a maximum. This corresponds to the optimal
percentage of investment in outreach, (that is, getting children into treatment) poptt ¼ 0:51
above which further investment does not yield payoff as the number of children in treatment
begin to go down. This is due to the fact that children with nodding syndrome need regular
follow-up visits for food supplements and an anticonvulsant to control the atonic seizures.
Without follow-up, there is no evident improvement, and this leads could lead to less children
enrolling for medication and supplements. The red curve (representing stunted children)
shows the point when R is minimum obtained when pt ¼ poptt : This is the point when pmin
f ¼
1 � poptt below which investment in follow-up does not prevent children from dropping out of
treatment. It is observed from the red curve that increase in pt corresponds to a reduced pf and
results in increase in stunted growth R. These results are obtained under the assumption that
healthcare is widely available, and symptomatic children enter treatment within one month of
displaying symptoms, to give the saturation level of outreach when at its most effective.
Using the value of poptt obtained from Fig 2, the model is simulated to study the behavior of
states I, T and R around this point for pt < poptt and pt > popt
t : Fig 3 shows the change in the
number of symptomatic children I, those enrolled in treatment T and the stunted or removed
R when pt = 0.1, 0.51, and 0.9. It is observed from the figure that the point poptt is obtained after
2 months. This implies that it gets harder to keep children in treatment after two successive
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 7 / 14
visits, and further investment without keeping those enrolled does not prevent stunted growth.
The number of children after two months of treatment with the different levels of investment
in outreach is approximately I(0.10) = 227, I(0.51) = 112, and I(0.90) = 97. We see that the
jump in the number of children on treatment from 10% to 51% investment is 115, while that
from 51% to 90% is 15. Similarly, T(0.10) = 556, T(0.51) = 716, and T(0.90) = 738; and R(0.10)
= 177, R(0.51) = 138, and R(0.90) = 132, with big jumps from T(0.10) to T(0.51) as compared
to T(0.51) to T(0.90). Thus when optimal investment in treatment is attained further input
does not yield significant benefits.
To obtain poptt , it was assumed that the treatment rate α is 1, while the rate of stunted devel-
opment after fall-out from treatment β is 0.1464. However, we can obtain different optimal
investment strategies for different sets of combinations of α and β to minimize stunted growth.
This is done by simulating the model for α, β 2 [0, 1] to determine when to invest in getting
children into treatment and when to switch to follow-up.
Fig 2. Simulation of the steady states with pt 2 [0, 1]. The red line is for the stunted while the blue line is for the children on treatment.
Note that maximum T (minimum R) is achieved at the point pt = 0.51. This gives poptt : The parameter values used are given in Table 1.
https://doi.org/10.1371/journal.pone.0172401.g002
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 8 / 14
Fig 4a and 4b shows the results of log(α) simulated against log(β) for optimal investment in
getting children into treatment pt and the respective number of children who develop stunted
growth R. It is observed from the figure that when the treatment rate α is large, it does not pay-
off to invest more in getting children in treatment and we should optimize follow-up. The
value of pt that minimizes stunted growth depends on α and β. For most of the α and β values,
pt is large indicating that investment in getting children into treatment does not prevent
stunted growth. These results are observed over a wide range of α and β although direct invest-
ment in getting children into treatment is optimal when α is small and β large.
Discussion and conclusion
Nodding syndrome has lasted a period of over five years in Pader and Lamwo districts in
northern Uganda killing mainly children between 5 and 15 years [2, 30, 31]. As a result, more
than 3,094 children are symptomatic and over 170 dead, in the war torn region with extreme
poverty. This paper provides a possible strategy of how limited resources can be utilized to
optimize control of the disease, first by getting children into treatment, and secondly, keeping
Fig 3. Simulation of the model for different values of pt 2 [0, 1], and initial conditions I = 1,000, T = 0, and R = 0. Parameter values used are in Table 1.
https://doi.org/10.1371/journal.pone.0172401.g003
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 9 / 14
them there. We hypothesize that getting children into treatment is necessary but keeping them
there is sufficient to prevent the affected children from stunted growth. We design a mathe-
matical model parameterized using data from Pader and Lamwo districts of Uganda, to pro-
vide a reliable optimal investment strategy that prevents stunted growth. The model is used to
show that if all children access treatment within a month of displaying symptoms, then up to
51% of available resources should be invested in getting children into treatment and not less
than 49% should be put in follow-up. Fig 2a shows that there is a maximum number of chil-
dren in treatment at the optimal point, suggesting that further investment would not be eco-
nomical. Fig 2b shows that as investment into medication is increased, the number of stunted
children reduce. At the point when optimal investment into getting children into treatment is
attained, we obtain the minimum number of stunted children; beyond this point, the number
of children that develop stunted growth begin to increase. This is the point when to switch
from further investment in treatment to investing into follow-up. Looking at the number of
symptomatic children after two months with no investment (pt, pf = 0,) gives 689. Without
medication, these will eventually develop stunted growth and die. If all investment is put in fol-
low-up (that is, pf = 1, pt = 0), there are 266 stunted growth which is a 61% reduction. However,
if 51% is invested in outreach (pt = 0.51, pf = 0.49), there is a reduction of 80% stunted growth.
From this we conclude that the best strategy is to invest about half in outreach, and half on fol-
low-up and predicts that the percentage number of stunted growth cases will be 80% less fol-
lowing this strategy compared to the simple strategy of all follow-up.
With the value of optimal pt obtained in Fig 2a, we simulated the model in Eq (1) to deter-
mine change if any, when values of pt < poptt and pt > popt
t are used. Fig 3a, 3b and 3c showed
the change in symptomatic, treated and stunted children with these values. Fig 3 indicates that
any investment strategy with a value of pt > poptt is not economical and does not yield any ben-
efits. This is the point at which a switch from outreach to follow-up should be made.
ure To further determine the best investment strategy, we simulated the steady states while
varying the rate α, at which treatment starts in an ideal situation, and the treatment fall-out
Fig 4. Simulation of the model for different values of α, β 2 [0, 1]. (a) Values of pt for all α and β combinations; (b) Corresponding values of R.
Parameter values used are in Table 1.
https://doi.org/10.1371/journal.pone.0172401.g004
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 10 / 14
rate β. Fig 4 postulates that when the symptomatic children access treatment immediately after
onset of symptoms and the fall-out rates are low, more investment should be put in getting
more children into treatment (Fig 4). If the children enroll in treatment immediately but the
fall-out rates are very high, then there should be little or no investment (0–20%) in getting chil-
dren into treatment; most investment (80–100%) should be utilized for follow-up. However, if
the children take very long to access treatment and the fall-out rates are very high, then up to
60% should be utilized to get more children into treatment and about 40% into follow-up.
When access to treatment is slow with low fall-outs, then all resources should be invested into
getting more children into treatment. It is important to note that the fall-out rates determined
the investment strategy for our model. Fast access to treatment coupled with high fall-out rates
call for more investment into follow-up (0–20%), while slow access to treatment with high fall-
out rates call for about 55% investment into treatment and 45% into follow-up. Fig 4 shows
that when both α and β are high, that is, fast access to treatment and high fall-out rates, it is not
optimal to further invest into getting children into treatment, since most of them are in treat-
ment. Therefore, it is more economical to invest in follow-up to reduce the high fall-out rates
and minimize the development of stunted growth.
This study provides the best way of utilizing limited resources to effectively control nod-
ding syndrome and minimize deaths and further development and stunted growth. The
results we obtained here show that prevention of stunted growth depends on two parameters;
how fast a child accesses medication α, and the probability of remaining on treatment β. In
most cases, when access to medication and fall-out rates are low, we should make greater
investment in getting more children into treatment than in follow-up, i.e optimal pt
remained high for all low values α, and more children developed stunted growth for all val-
ues of β and α< 11.85%. In most cases, optimal investment into getting children into treat-
ment pt is 100%, except for the instances when fall-out rate β is very high (>4), and access to
treatment rate α is very small (<7.29%). The results show that there is a limit to how much
you can invest in getting children in treatment and keep them there without developing
stunted growth. In the nodding case discussed here, local council representatives visited
homes with suspected children and urged their next of kins to take them to screening and
outreach centers for medication and management information. The key study limitation was
not knowing the onset of symptoms of nodding syndrome such that initial dates were taken
to be when parents suspected their children of having nodding disease after repeated seizures
and nodding. The approach we use here can help control nodding children from developing
stunted growth especially when faced with limited information and show what the benefits
of outreach and follow-up against nodding syndrome could have been. Moreover our con-
clusions are robust against the most uncertain parameters in our model such as δT and δR.
Doubling or halving these values preserves the advantage of the mixed strategy over the sin-
gle strategies. However, other diseases that have been neglected due to limited resources
could be controlled in a similar way.
Supporting information
S1 Fig.
(PDF)
S2 Fig.
(PDF)
S3 Fig.
(PDF)
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 11 / 14
S4 Fig.
(PDF)
S1 File.
(PDF)
S2 File.
(PDF)
S3 File.
(PDF)
S4 File.
(PDF)
S5 File.
(PDF)
Acknowledgments
BN acknowledges funding from the Faculty for the Future and the Center for Interdisciplinary
Mathematics, Mathematics Institution, Uppsala University, Sweden. BN thanks the Ministry
of Health Uganda for the data. The authors extend gratitude to the constructive comments
from the reviewers.
Author Contributions
Conceptualization: BN DS.
Data curation: BN DS.
Formal analysis: BN DS.
Funding acquisition: BN DS.
Investigation: BN DS.
Methodology: BN DS.
Project administration: BN DS.
Resources: BN DS.
Software: BN DS.
Supervision: BN DS.
Validation: BN DS.
Visualization: BN DS.
Writing – original draft: BN DS.
Writing – review & editing: BN DS.
References1. BBC News. http://www.bbc.co.uk/news/world-africa-17589445, 2012. Accessed June 23 2012.
2. Lacy M. Nodding disease: mystery of southern Sudan. Lancet Neurology, 2014; (2)12:714. https://doi.
org/10.1016/S1474-4422(03)00599-4
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 12 / 14
3. Wadman M. African outbreak stumps experts. Nature. 2011. https://doi.org/10.1038/475148a
4. UGANDA: Nodding disease or ”river epilepsy?” http://www.irinnews.org/Report/85646/UGANDA-
Nodding-disease-or-river-epilepsy. Accessed June 23 2012.
5. Nodding disease in East Africa, CNN. http://edition.cnn.com/video/#/video/world/2011/06/05/africa.
dowell.nodding.disease.cnn?hpt=hp_mid. Accessed March 06 2012.
6. Ministry of Health Uganda: Statement by the minister of health. www.health.go.ug/docs/STATEMENT.
pdf. Accessed January 15 2013.
7. Harding A. Nodding disease hits Sudan. BBC News 23 September 2003. http://news.bbc.co.uk/2/hi/
africa/3133440.stm. Accessed March 06 2012.
8. Centers for Disease Control and Prevention (CDC). Nodding Syndrome. http://www.cdc.gov/
globalhealth/noddingsyndrome/. Accessed December 27 2012.
9. Ross E. Bizarre Illness Terrifies Sudanese-’Nodding Disease’ Victims Suffer Seizures, Retardation,
Death, CBS News, Jan. 28, 2004. http://www.cbsnews.com/stories/2004/01/28/health/main596493.
shtml. Accessed March 06 2012.
10. Ross E. Sudan A Hotbed Of Exotic Diseases”. CBS News (Rumbek, Sudan), (3 February 2004). http://
web.archive.org/web/20040218105940/ http://www.cbsnews.com/stories/2004/02/03/health/
main597751.shtml. Accessed March 06 2012.
11. Abraham C. Mysterious nodding syndrome spreading through Uganda. New Scientist. 2011. http://
www.newscientist.com/article/dn21316-mysterious-nodding-syndrome-spreading-through-uganda.
html.
12. The Monitor, Uganda’s daily newspaper. Uganda: Nodding disease and malnutrition. July 15 2005.
http://crofsblogs.typepad.com/h5n1/2012/07/uganda-nodding-disease-and-malnutrition.html.
Accessed January 21 2013.
13. Butagira T. Nodding disease: Crystals found in victims’ brains. http://www.monitor.co.ug/News/
National/Nodding-disease–Crystals-found-in-victims–brains/-/688334/2403276/-/item/1/-/9rxioj/-/index.
html. 2014. Accessed August 26 2014.
14. Makumbi C. Parents neglect nodding children. http://www.monitor.co.ug/News/National/Parents-
neglect-nodding-children/-/688334/2412946/-/m8ixjxz/-/index.html. 2014. Accessed August 27 2014.
15. Ocungi J. School bars nodding child from attending class. http://www.monitor.co.ug/News/National/
School-bars-nodding-child-from-attending-class/-/688334/2406320/-/dusli5z/-/index.html. 2014.
Accessed August 27 2014.
16. Kamien MI, Schwarz NL. Dynamic optimization: the calculus of variations and optimal control. North
Holland, Amsterdam. 1991.
17. Kar T, Ghosh B. Sustainability and optimal control of an exploited prey predator system through provi-
sion of alternative food to predator. Biosystems. 2012; 220–232. https://doi.org/10.1016/j.biosystems.
2012.02.003 PMID: 22370042
18. Fleming WH, Rishel RW. Deterministic and stochastic optimal control. Springer Varlag, New York.
1975.
19. Pontryagin LS, Boltyanskii VG, Gamkrelidze RV, Mishechenko EF. The Mathematical Theory of Opti-
mal Processes. Wiley, New York. 1962.
20. Garcia LS, David AB. Diagnostic Medical Parasitology 3rd ed. Washington: ASM Press. 1997.
21. Castillo-Chavez C, Feng Z, Huang W. On the computation of R0 and its role on global stability. In: Math-
ematical approaches for emerging and re emerging infectious diseases. An Introduction. Castillo-Cha-
vez C, Blower S, van den Driessche P, Kirschner D, Yakubu AA. (Eds.) Springer-Verlag, New York.
2002; 229–250.
22. Castillo-Chavez C, Song B. Dynamical Models of Tuberculosis and their Applications. Mathematical
Biosciences and Engineering. 2004; 1(2): 361–404. https://doi.org/10.3934/mbe.2004.1.361 PMID:
20369977
23. Chester P, Jung RC, Wayne E. Clinical Parasitology 9th ed. Beaver, Cupp. Philadelphia: Lea and
Febiger. 1984.
24. Chitnis N, Hyman JM, Cushing JM. Determining important parameters in the spread of Malaria through
the sensitivity analysis of a mathematical model. Bulletin of mathematical biology. 2008, 70(5):1272–
1296. https://doi.org/10.1007/s11538-008-9299-0 PMID: 18293044
25. Osei-Atweneboana MY, Eng JKL, Boakye DA, Gyapong JO, Prichard RK. Prevalence and Intensity of
Onchocerca volvulus infection and efficacy of ivermectin in endemic communities in Ghana: a two-
phase epidemiological study. Lancet. 2007; 369(9578): 2021–9. https://doi.org/10.1016/S0140-6736
(07)60942-8 PMID: 17574093
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 13 / 14
26. Nannyonga B, Sumpter DJT, Mugisha JYT, Luboobi LS. The Dynamics, Causes and Possible Preven-
tion of Hepatitis E Outbreaks. PLoS ONE. 2012; 7(7): e41135. https://doi.org/10.1371/journal.pone.
0041135 PMID: 22911752
27. Kobayashi M, Carpenter TE, Dickey DF, Howitt RE. A dynamic, optimal disease control model for foot-
and-mouth-disease: II. Model results and policy implications. Preventive Veterinary Medicine. 2007; 79:
274–286. https://doi.org/10.1016/j.prevetmed.2007.01.001 PMID: 17280730
28. Rodrigues HS. Optimal Control and Numerical Optimization Applied to Epidemiological Models. PhD
Thesis, Doctoral Programme in Mathematics and Applications (PDMA), University of Aveiro and Univer-
sity of Minho. 2012.
29. Deng Y, Shen S, Vorobeychik Y. Optimization methods for decision making in disease prevention and
epidemic control. Mathematical Biosciences. 2013; 246(1): 213–227. https://doi.org/10.1016/j.mbs.
2013.09.007 PMID: 24121040
30. Foltz JL, Makumbi I, Sejvar JJ, Malimbo M, Ndyomugyenyi R, Atai-Omoruto AD, Alexander LN, Abang
B, Melstrom P, Kakooza AM, Olara D, Downing RG, Nutman TB, Dowell SF, Lwamafa DKW. An Epide-
miologic Investigation of Potential Risk Factors for Nodding Syndrome in Kitgum District, Uganda.
PLoS ONE. 2013; 2013; 8(6): e66419. https://doi.org/10.1371/journal.pone.0066419 PMID: 23823012
31. Sejvar JJ, Kakooza AM, Foltz JL, Makumbi I, Atai-Omoruto AD, Malimbo M, Ndyomugyenyi R, Alexan-
der LN, Abang B, Downing RG, Ehrenberg A, Guilliams K, Helmers S, Melstrom P, Olara D, Perlman S,
Ratto J, Trevathan E, Winkler AS, Dowell SF, Lwamafa DKW. Clinical, neurological, and electrophysio-
logical features of nodding syndrome in Kitgum, Uganda: an observational case series. Lancet Neurol.
2013; 12: 166–74. https://doi.org/10.1016/S1474-4422(12)70321-6 PMID: 23305742
32. Districts of Uganda. http://en.wikipedia.org/wiki/Districts_of_Uganda. Accessed September 20 2014.
33. Dugdale DC, Epstein R, Pantilat SP. Time and the Patient? Physician Relationship. J Gen Intern Med.
1999. 14(Suppl 1): S34–S40. https://doi.org/10.1046/j.1525-1497.1999.00263.x PMID: 9933493
34. Roland MO, Bartholomew J, Courtenay MJ, Morris RW, Morrell DC. The “five minute” consultation:
effect of time constraint on verbal communication. Br Med J (Clin Res Ed). 1986; 292(6524):874–6.
https://doi.org/10.1136/bmj.292.6524.874
Cost allocation of limited resources to control nodding syndrome
PLOS ONE | https://doi.org/10.1371/journal.pone.0172401 March 13, 2017 14 / 14